JP2740506B2 - Image recognition method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognition method, and more particularly to an image recognition method for recognizing a character symbol corresponding to a pattern from a character symbol pattern in image data. 2. Description of the Related Art Conventionally, there is a character recognition device that recognizes a regular character from a character symbol pattern in image data input by a document reading device. This type of device is based on the premise that only a character symbol pattern exists in image data, and cannot be recognized when, for example, a character symbol pattern and a pie chart pattern are mixed. . Furthermore, in this type of character recognition device, recognition processing is performed from the input character pattern itself. That is, since the recognition process is not assisted based on the state or environment in which the character pattern to be recognized is placed, the recognition efficiency is still not sufficient. [Problems to be Solved by the Invention] The present invention has been made in view of the above conventional technology,
An object of the present invention is to provide an image recognition method that improves the character recognition rate by referring to not only the character recognition processing but also the area ratio of each item in the graph for the numerical characters of each item in the graph. [Means for Solving the Problems] The image recognition method of the present invention for solving the problems includes the following steps. That is, an image recognition method for narrowing down candidates from a group of candidates obtained by character recognition of numerical characters indicating a ratio in a graph, and storing image data of a graph including a character indicating a trigger to be recognized. Thinning the stored image data, extracting a continuous point sequence forming a line segment forming the graph from the thinned image data, and extracting vector information based on the continuous point sequence. Generating, calculating the area ratio of each item of the graph according to the generated vector information, and using the obtained area ratio, a candidate is selected from a candidate group obtained by recognizing numerical characters of each item in the graph. To narrow down. Embodiment An embodiment according to the present invention will be described below in detail with reference to the accompanying drawings. FIG. 1 is a block diagram of the information recognition apparatus of the present embodiment. In the figure, reference numeral 1 denotes an original, which is the original 1 in this embodiment.
It is assumed that text information ("x" means a character) and a pie graph have already been written as shown in FIG. 2 (a).
Reference numeral 2 denotes an image reading device that optically reads the document 1.
Reference numeral 3 denotes a buffer memory for storing read image data. Reference numeral 4 denotes a pie chart recognizing unit for recognizing a pie chart in the image data developed in the buffer memory 3, the details of which will be described later. Reference numeral 5 denotes a character recognition unit for recognizing a character / symbol pattern in the image data developed in the buffer memory 3 based on the information in the character font ROM 6 and the information obtained by the pie graph recognition unit 4. For the character pattern recognized by the character recognition unit 5, a regular character pattern is read from the character font ROM 6, and a VRAM7 which is a RAM for display is read.
Process to write to. The various patterns stored in the VRAM 7 are also displayed on the CRT 9 by the graphic controller GC8. Further, reference numeral 10 denotes a CPU for controlling the entire apparatus, a ROM 10a for storing a control program (flow charts in FIGS. 4 and 6) for controlling the entire apparatus, and a RAM 10b used as a work area of the CPU 10. It has. [Explanation of Pie Graph Recognition Unit (FIGS. 2A to 2C, 3rd
FIG.)] In the configuration described above, if the image on the document 1 is as shown in FIG.
FIG. 2A also shows the image data stored in the buffer memory 3 via. When the image data is stored in the buffer memory 3, the pie chart recognition unit 4 reads the image from the buffer memory 3 and thins the image data (FIG. 2 (b)).
By vectorizing only the point sequence, only the straight and circular patterns shown in FIG. 2 (c) are extracted.
This extracted image may be expanded in VRAM7,
In the present embodiment, an unillustrated
Assume that it is expanded to RAM. Now, when recognizing the pie chart thus obtained, in the present embodiment, as shown in FIG. 3, the angle θ _{A in} the area _A , the angle θ _{B in the} area _B, and the angle θ _{C in the} area _C.
From this, the angle ratio (equivalent to the area ratio) to the entire circle in each area is calculated. Since θ _A + θ _B + θ _C = 360 °, the angle ratio of each area to the entire circle (tentatively referred to as σ _A , σ _B , σ _C )
Is calculated by the following equation. That is, FIG. 4 shows an example of σ _A , σ _B , and σ _C calculated using the above formula. In addition, these σ _A , σ _B , and σ _C mean the numerical characters themselves representing the percentage of each area, that is, the frequency ratio in each area in the pie chart in the document 1. Therefore, when the character recognizing unit 5 recognizes numerical characters representing the frequency ratio in the pie chart (these characters may be smaller than other characters, and a large number of recognition candidates may be generated). ) Can be used as an element for selecting one from a plurality of candidate numeric characters. [Explanation of Recognition Processing (FIGS. 5 to 7)] Based on the above pie graph recognition processing, the flow of the processing of this embodiment will be described with reference to the flowchart shown in FIG. First, in step S1, the original image 1 is read and stored in the buffer memory 3. Next, the process is divided into a pie graph recognition process in step S2 and a character segmentation process in step S4. First, step S2 will be described. In step S2, the image data stored in the buffer memory 3 is thinned, vectorized, and a pie chart excluding a character pattern is recognized. Next, in step S3, the frequency ratio for each area in the pie graph is calculated. At step S4, only character patterns in the image stored in the buffer memory 3 are cut out (step S4) in parallel with steps S2 and S3, and normal character recognition processing is executed (step S5). At this time, characters in the pie chart and ratio characters are also recognized. Note that the processing in steps S4 and S5 does not necessarily need to be performed in parallel with steps S2 and S3, and may be performed after steps S2 and S3. In the case of character segmentation, for example, when a pie chart is recognized in step S2, image data other than the pie graph (which eventually becomes a character pattern) may be character-recognized. Now, in step S6, based on the result of the character recognition process (step S5), each area obtained in step S3 is selected in order to select one from the recognition candidate group of the frequency ratio for each area in the pie chart. The best recognition candidate is selected with reference to the frequency ratio value (σ _A , σ _B , σ _C ) for each. Next, a regular character pattern corresponding to the character pattern recognized in step S7 is read from the character font ROM 6, expanded in the VRAM 7, and displayed. At this time, the pie chart pattern itself is also transferred to the VRAM 7 (reduced and transferred if necessary). For example, the recognition candidate characters of the frequency ratio values of the areas A, B, and C in the pie graph recognized by the character recognition unit 5 in step S5 are as shown in FIG. The processing in step S6 when the frequency ratio values obtained for each area are as shown in FIG. 4 (b) is as shown in the flowchart of FIG. In step S70, the candidate value “55” having the smallest difference from σ _A (= 54.5) of area A is determined as the recognition result.
Hereinafter, “3” is set in step S71, and “42” is set in step S72.
Is the recognition result. As described above, even in a document in which a pie chart and text information are mixed, only a character pattern can be cut out and recognized. In addition, since the frequency ratio value in the pie chart can be predicted to some extent from the pie chart itself, based on this prediction,
The efficiency of character recognition of the frequency ratio value will be significantly improved. [Effects of the Invention] As described above, according to the present invention, character recognition is performed by referring not only to the character recognition processing but also to the area ratio of each item in the graph for numerical characters of each item in the graph. The rate can be improved.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram of a character recognition device according to the present embodiment, FIG. 2 (a) is a diagram showing an original image, and FIG. 2 (b) is a diagram showing thinned image data. FIG. 2 (c) is a diagram showing an image obtained by extracting a pie graph from the image of FIG. 2 (b). FIG. 3 is a diagram for explaining calculation of a frequency ratio value by using a pie graph. FIG. 4 is a diagram showing an example of calculation of each frequency value in a pie chart. FIG. 5 is an operation flowchart of a character recognition process in this embodiment. FIG. FIG. 7 is a flowchart showing a procedure for selecting a frequency ratio value to be a recognition result. In the figure, 1 ... original, 2 ... image reading device, 3 ... buffer memory, 4 ... pie graph recognition unit, 5 ... character recognition unit, 6 ... character font ROM, 7 ... VRAM, 8 ... GC,
9 ... CRT, 10 ... CPU, 10a ... ROM, 10b ... RAM.

Claims (1)

(57) [Claims] An image recognition method for narrowing down candidates from a group of candidates obtained by character recognition of numerical characters indicating a ratio in a graph. The obtained image data is thinned, a continuous point sequence forming a line segment forming the graph is extracted from the thinned image data, and vector information is generated based on the continuous point sequence. Calculating the area ratio of each item of the graph according to the generated vector information, and narrowing down candidates from a group of candidates obtained by recognizing numerical characters of each item in the graph using the obtained area ratio. An image recognition method characterized in that: